Fix arm64 macOS build with Homebrew

[audriusbuika]

Problem

On Apple Silicon, ./src/Build/build_veracrypt_macosx.sh -b fails to build:

clang: error: unsupported option '-mssse3' for target 'arm64-apple-darwin24.6.0'
clang: error: unsupported option '-msse4.1' for target 'arm64-apple-darwin24.6.0'
clang: error: unsupported option '-msha'   for target 'arm64-apple-darwin24.6.0'
make[1]: *** [../Crypto/Sha2Intel.oshani] Error 1

Once the flag issue is worked around, the link step then fails with missing
arm64 symbols: _aes_encrypt, _aes_hw_cpu_*, and
_sha256_compress_digest_armv8.

Root Cause

Two separate issues surface together on arm64:

1. The pattern rules in src/Build/Include/Makefile.inc (%.oshani,
   %.osse41, %.ossse3, %.oaesni, %.oavx2) always append x86-only
   flags such as -mssse3 -msse4.1 -msha. clang rejects those flags when
   targeting arm64, even though the source bodies are already guarded by
   CRYPTOPP_BOOL_X86/X64 and would compile to empty objects.

2. src/Volume/Volume.make has no arm64 object list for local MacOSX
   builds. The build script sets COMPILE_ASM=false on arm64 hosts (yasm
   can't assemble x86 asm for an arm64-only build), which leaves the
   MacOSX branch empty, and the generic arm64 branch below it is never
   reached. As a result, no AES or SHA-256 backend ends up in the link.

Fix

src/Build/Include/Makefile.inc — move the x86 feature flags into
variables that are empty when CPU_ARCH=arm64. x86/x64 builds keep the
same flags they had before; arm64 compiles the same sources without
them, and the existing #ifdef guards take care of the rest.

src/Volume/Volume.make —

1. Add an else ifeq "$(CPU_ARCH)" "arm64" branch inside the MacOSX
   block that pulls in Aes_hw_armv8.oarmv8crypto, Aescrypt.o, and
   sha256_armv8.oarmv8crypto — the symbols the linker was missing.
2. On arm64, compile Sha2Intel, blake2s_SSE41, blake2s_SSSE3, and
   opt_avx2 as plain .o files so they skip the x86-flag pattern rules.

Verification

The build now completes on arm64 (macOS 24.6, SDK 26.2) and produces a
working src/Main/VeraCrypt binary. x86_64 and universal-binary release
builds are untouched — the new paths only activate when CPU_ARCH=arm64.

[idrassi]

Thanks for investigating the Apple Silicon/Homebrew build failure.

The local arm64 build issue is real, but I do not think this PR cannot be merged as-is because the CPU_ARCH=arm64 conditions also affect the normal non-local macOS release build on Apple Silicon.

On Apple Silicon, CPU_ARCH is derived from uname -m, so it is arm64 even when the build is not local-only. In that non-local path, src/Makefile still builds a universal x86_64 + arm64 binary. With this PR, src/Volume/Volume.make therefore compiles blake2s_SSE41, blake2s_SSSE3, Sha2Intel and Argon2/src/opt_avx2 as plain .o files for the universal build instead of using the feature-specific .osse41, .ossse3, .oshani, and .oavx2 rules.

The opt_avx2 case is a functional regression: without -mavx2, __AVX2__ is not defined, so fill_segment_avx2 compiles to the stub returning ARGON2_INCORRECT_PARAMETER. The x86 runtime dispatch calls this function when HasSAVX2 is true, so the x86_64 slice of a universal macOS build produced on Apple Silicon can fail on AVX2-capable Intel machines.

Please scope the plain-object fallback to the local arm64 build only, for example by keying it on LOCAL_DEVELOPMENT_BUILD=true and CPU_ARCH=arm64 rather than on CPU_ARCH=arm64 alone. The global Makefile.inc flag stripping should either be avoided or constrained by the same condition.

[idrassi]

This condition is too broad. CPU_ARCH=arm64 is also true for the normal non-local macOS build on Apple Silicon, which still produces a universal x86_64 + arm64 binary. In that path this block makes the x86_64 slice compile Argon2/src/opt_avx2.c as plain .o instead of .oavx2, so __AVX2__ is not defined and fill_segment_avx2() becomes the stub that returns ARGON2_INCORRECT_PARAMETER. Please scope this fallback to the local arm64-only case, e.g. LOCAL_DEVELOPMENT_BUILD=true plus CPU_ARCH=arm64, not CPU_ARCH=arm64 alone.

[idrassi]

This has the same scoping issue as the Volume.make change: CPU_ARCH=arm64 is true on Apple Silicon even for the normal non-local macOS build, which still targets a universal x86_64 + arm64 binary.
For that universal path, the x86_64 slice still needs these feature flags. In particular, clearing X86_SHANI_FLAGS makes Sha2Intel.oshani compile without -msha: with Apple clang this fails because _mm_sha256msg1_epu32 requires the sha target feature.
Please either leave these pattern rules unchanged and avoid selecting the x86-specific suffix objects for local arm64-only builds, or gate this flag clearing on the local arm64-only case, e.g. LOCAL_DEVELOPMENT_BUILD=true plus CPU_ARCH=arm64, not CPU_ARCH=arm64 alone.